← Back to live feed1 story
Research led by Nick Cammarata at Goodfire found that the vast majority of neurons in trillion-parameter-scale models are legible, enabling direct tracing of circuits in systems like Kimi K3. Audits of Qwen3-32B showed roughly 90% of neuron poles serve a single main purpose, reversing earlier estimates from Cammarata's OpenAI work that only 2% of neurons were legible. The findings may transform AI alignment into a normal science and increase the likelihood of achieving mechanistic interpretability before superintelligence.
Sign in to suggest
Key sources
- SOURCE@nickcammarata“the vast majority in models up to trillions of parameters are legible, and their circuits beautiful”x.com
- SUPPORT@tszzl“i think this may be remembered as a major moment in scientific history. news so good even eliezer will have to lower his pdoom”x.com
- SUPPORT@andrewcurran_“I had long since given up on reaching any real level of interpretability before superintelligence arrived”x.com
- SOURCEhuggingnewshuggingnews.com
- SOURCEtechmeme.com